Professional Documents
Culture Documents
Heuristic Models For Multiparameter Identification of Hydrodynamic Objects of Analysis
Heuristic Models For Multiparameter Identification of Hydrodynamic Objects of Analysis
Heuristic Models For Multiparameter Identification of Hydrodynamic Objects of Analysis
COMPUTER SCIENCES
https://cajmtcs.centralasianstudies.org
The solution of many problems of statistical data processing is associated with the generalization and
identification of patterns in the processes under study based on observations of their real flows or physical
experiments. These processes are always influenced by many random and discrete factors. In a physical
experiment, statistical data are collected, which are usually incomplete. In such cases, the only research
method is the mathematical analysis of the information received, where heuristic self-organization methods
are widely used [1]. One of them is the method of group accounting for an argument (GAM), which is based
on a multi-row selection of particular models with successive complications.
A.G. Ivakhnenko [1] gives a list of programs based on GAM algorithms. The logical substantiation and a
fairly accurate presentation of the principle of self-organization contributed to the deepening of these
studies.
This article is devoted to the development of algorithms for parametric identification of the qualitative and
dynamic states of the system, respectively.
Determination of qualitative estimates of geophysical parameters under various artificial impacts on the
reservoir system is one of the most important issues in the analysis of oil and gas field development
processes. To identify the information content of features, a generalized parametric identification algorithm
has been developed to study the dynamic states of the system based on the principle of “Heuristic self-
organization”.
The input data of the object of study here are the results of a physical experiment, which are accepted as
training information, and the information obtained with the help of the constructed model is as a verification
one.
To select the optimal object model, the algorithm provides for checking the standard deviations using the
“left corner” principle.
Main part
Let us consider the construction of heuristic models of multipara meter objects based on the results of a few
physical experiments. Let some information about the studied process be collected from observations with
volume , which is the starting point for building the model. This volume of information is divided into
three parts: training 0 , verification n and examining э . In contrast to the algorithms of the method of
group accounting for an argument (GAM), this separation is not carried out in a unique way. The model built
by the GMDH method is optimized by the ranking of the initial information . So there is a third possibility
of choosing the optimal model. For this, root mean square errors are used. э , calculated by the optimal
model of each ranking of the initial information .
Thus, the task of constructing a mathematical model of multi-parameter processes based on the MGAA
reduces to the following: incomplete information is given , consisting of system parameter values
xi xij , i 1, n , corresponding to the studied parameter y j , j 1, R and parameters at the predictive point
xni . Required to define a dependency function y f ( x1 , x2 , , xn ) and its predicted value yn ( n -number
of measured parameters in the system, R - number of observations).
Stages of building a three-criteria selection model.
1. Transition from dimensional to dimensionless variable formulas
*
/
xij xij xi max , i 1, N ,
(1)
y* y
ij ij /
ymax , j 1, N ,
xi max , ymax - maximum values. This stage makes it possible to calculate the values necessary for constructing
the model, regardless of the setting of the system of parameter dimensions, therefore, all parameters can be
represented in the dimensions accepted within the studied limits. This makes it easier to prepare the initial
information.
2. Pre-processing of initial information. We calculate the range of parameters in the response function, the
mean value of the variance and the standard deviation using the formulas.
Px xi max xi min ,
Py ymax ymin ,
R
xi xij / R, (2)
j 1
R
Si2 ( xij xi ) 2 / ( R 1), Si Si2 , i 1, N .
j 1
3. Selection of the examining sequence from the initial information. To do this, we calculate the distance
between the point and the steel points of the definition using the formula
N
i
i 1
( xij xni )2 , j 1, R (3)
4. Separation
u on the o and n with volume R1 and R2 . For this point from u , located in
ascending distance from the forecast point, are collected in various combinations and are formed o and
n . For example, o consists of points with even numbers, а n - odd. Vice versa, n consist of even,
o - from the odd ones. The first two in o , the third is in n etc. Can arrange M 1 combinations.
x
5. Pairing k and xe parameters and construction of particular descriptions. Dependencies can be taken
from support functions
ys a0 a1 xk a2 xe a3 xk xl (4)
or
ys a0 a1 xk a2 xl a3 xk xl a4 xk2 a5 xl2 ,
k 1, N 1, l 2, N , l 2, M 2 ,
Dependencies of the form (4) are constructed by the least squares method in o . The first selection criterion
is used
1 R1
y j ysj*
2
min .
R1 j 1
Here ysj* calculation according to dependence (4) of the value of the studied parameter y j .
6. Calculation of quantities
s2
1 R1
y j ysj*
2
R2 j 1
and ordering particular models in ascending order s2
12 22 M2 .
2
Here required calculation accuracy; 1*2 mean square error values from the previous selection
corresponding to the best of particular descriptions.
*2
Condition (5) corresponds to the “left corner” principle. After fulfilling the condition 12 1 . The next
selection does not improve the model; in the general case, condition (5) for this statistic is also satisfied for
all subsequent selections, i.e. further improvement of the model is not possible.
This is how the only optimal model is determined that satisfies the criterion of minimum mean square error
and corresponding to M - th ranking of the initial information uc .
7. Calculation of the value of the root-mean-square error for the examining sequence in the optimal model
using the formula
2
1 M1
m2
M 1 j 1
y j yib ,
where yib - calculated value corresponding to the table y j and calculated by the optimal model.
8. Repeat steps 5-7 for another ranking. This determines the set of the best models corresponding to the
ranking options.
9. Model corresponding to the minimum of the set ̂ , is considered the best and its accuracy is determined
by the formula
̂
Thus, the best model of multiparametric processes based on the principle of self-organization is built. It is
practically impossible to get the model form, as in [1], because the addition of ranking options will again
increase the number of terms in the polynomial. Apparently, this is not necessary. If you want to get the
numerical value of the forecast, this can be achieved by calling the function procedures.
When using the above algorithm, its features should be taken into account. If one of the pairs of arguments
xk , xl takes uniform values, i.e. they can be represented as
xk jh1 , xl jh2 , j 1, R1 ,
j j
then the determinant of the normal system of equations is equal to zero. In this case, it is impossible to
obtain solutions to the normal system of equations and, consequently, to build a model for this pair. We
exclude the model corresponding to this case from the set of models and proceed to the next pair of
parameters. Therefore, the total number of models in each row of selections may be less than CN2 .
Consider the problem of determining the parameters of the system if the value of the response function is
given. Let the best model of the system be built, it is required to determine the values of the parameters
( x10 , x20 , , xn0 ) under which the function f ( x10 , x20 , , xn0 ) f 0 . It is known that this problem reduces to
finding the minimum of the functional
F ( x1 , x2 , , xn , y0 ) f ( x1 , x2 , , xn ) y0
where K н - initial air permeability of the core to saturation with solution; K нт - permeability of the core with
washed crust; Т нс - duration of core saturation with solution; V - the volume of the passed filtrate; -
solution viscosity; - specific gravity of the solution; B - water yield in 30 min. solution; Cнс - static shear
stress.
Table 2
K нт Т пр B Cнс Kн
168.3 2,5 117 1,02 7 18 725,5
150,4 3 62 1,02 38 0 553,9
507,8 1,75 36 1,04 8 84 946,01
428,2 1,66 50 1,05 8 12 814,56
128,2 3,5 28 1,062 9 0 539,5
287 2,75 28 1,07 8 0 584,4
204,9 4 56 1,12 10 24 827,37
448,1 2,33 38 1,13 18 20 857,33
229 2,66 28 1,14 12 9 609,53
198,3 2,75 28 1,13 32,5 0 416,66
2. The process of displacement of the filtrate by gas. In this case, the permeability of the reservoir can be
slightly improved. The dependence is considered
K n K n ( K нм , Т пр , , , B, Cнс ) (7)
The change in rock permeability after gas blowing is characterized by the following data (Table 4):
Table 4
Kн K нb ( K н K нb ) / K нb
725,5 703,86 0,0298
553,8 548,69 0,0092
946,01 931,18 0,01577
814,56 827,46 -0,0158
539,5 506,39 0,0614
584,4 567,23 0,0294
827,37 828,03 -0,0008
857,53 874,02 -0,0198
609,53 634,77 -0,0411
416,66 445,15 -0,0684
The optimal model was obtained for the first stage after 27, for the second after 6 rows of selections. The
mean square errors are 0.0796 and 0.00034. The reason for stopping the selection of models was the “left
corner” principle for both stages, i.e. improvement of the mean square error in the future is impossible. The
relative error ranges from 1 to 8%. Similar results were obtained with other data of the problem.
Naturally, in the cases described, it is fundamentally impossible to bring the model into an analytical
connection of the response function of the studied parameters. Such a model would include at least 28 terms
of varying degrees of variables.
Thus, the proposed method can be used to build an optimal model for changing the filtration properties of
reservoirs when flushing fluids with different properties move through them [4].
A good agreement between the experimental and calculated data shows the acceptability and effectiveness of
the proposed parametric identification algorithm when developing a formulation of lightweight drilling
fluids with and without solids to improve the technological and technical and economic development
indicators when drilling wells in difficult conditions, as well as when killing gas wells on the period of
production of repair work at the fields
References
1. Ivakhnenko A.G., Zaichenko Yu.P., Dimitrov V.D. Decision making based on self-organization. - M.:
Science. - 1976. - 280 p.
2. Rastrigin L.A. Statistical search methods. - M.: Nauka. - 1969. - 376 p.
3. Rastrigin L.A. Extreme control systems. - M.: Science. 1974. - 630 p.
4. Suvonov O.O. Mathematical model of control of a hydrodynamic object with distributed parameters (on
the example of oil fields). Central Asian journal of mathematical theory and computer sciences
http://cajmtcs.centralasianstudies.org/index.php/CAJMTCS Volume: 03 Issue: 04 | Apr 2022 ISSN:
2660-5309. 5 pages.